| hyper-parameters | |
|---|---|
| BPR | learning_rate=0.001 weight_decay=0.00001 val_interval={rating: "[3,inf)"} neg_sample_num={"uniform": 1} |
| NCF | learning_rate=0.001 mlp_hidden_size=[128,64] dropout_prob=0.2 weight_decay: 0.000001 threshold={rating: 3.0} |
| LightGCN | learning_rate=0.001 n_layers=2 reg_weight=0.0001 threshold: {rating: 3.0} val_interval={rating: "[3,inf)"} neg_sample_num={"uniform": 1} |
| PMF | weight_decay=0.0001 val_interval={rating: "[3,inf)"} neg_sample_num={uniform: 1} |
| BiasedMF | learning_rate=0.001 weight_decay: 0.0001 val_interval={rating: "[3,inf)"} neg_sample_num={uniform: 1} |
| DMF | learning_rate=0.001 num_layers=3 weight_decay=0.001 val_interval={rating: "[3,inf)"} neg_sample_num={uniform: 1} |
| FOCF | learning_rate=0.0001 fair_weight=1 weight_decay=0.001 fair_objective is selected from [none,value,absolute,under,over,nonparity] according specific objective threshold={rating: 3.0} |
| PFCN_BiasedMF | learning_rate=0.001 dis_hidden_size_list=[128, 256, 128, 128, 64, 32] activation=leakyrelu filter_mode is selected from [none,sm,cm] according specific filter mode dis_dropout=0.3 train_epoch_interval=5 weight_decay=0.0001 dis_weight=10 val_interval={rating: "[3,inf)"} neg_sample_num={uniform: 1} |
| PFCN_DMF | learning_rate=0.001 num_layers=3 dis_hidden_size_list=[128, 256, 128, 128, 64, 32] mlp_activation=relu dis_activation=leakyrelu filter_mode is selected from [none,sm,cm] according specific filter mode mlp_dropout: 0.2 dis_dropout=0.3 train_epoch_interval: 5 weight_decay=0.001 dis_weight=10 val_interval={rating: "[3,inf)"} neg_sample_num={uniform: 1} |
| PFCN_MLP | learning_rate=0.001 mlp_hidden_size_list=[64, 32, 16] dis_hidden_size_list=[128, 256, 128, 128, 64, 32] activation: leakyrelu filter_mode is selected from [none,sm,cm] according specific filter mode dropout=0.2 dis_dropout=0.3 train_epoch_interval=5 weight_decay=0.0001 dis_weight=10.0 val_interval={rating: "[3,inf)"} neg_sample_num={uniform: 1} |
| PFCN_PMF | learning_rate=0.001 dis_hidden_size_list=[128, 256, 128, 128, 64, 32] activation=leakyrelu filter_mode is selected from [none,sm,cm] according specific filter mode dis_dropout=0.3 train_epoch_interval=5 weight_decay=0.0001 dis_weight=10 val_interval={rating: "[3,inf)"} neg_sample_num={uniform: 1} |
| FairGo_GCN(WAP) | pretrain_model_file_path= Your GCN Pretrain File Path aggr_method: 'WAP' n_layers=2 dis_hidden_size_list=[16,8,4] filter_hidden_size_list=[128,64] activation=leakyrelu fair_weight=0.1 gcn_n_layers=2 hidden_channels=32 gcn_dropout=0.2 gcn_act: relu train_epoch_interval=5 weight_decay=0.0001 threshold={rating: 3.0} |
| FairGo_GCN(LVA) | pretrain_model_file_path= Your GCN Pretrain File Path aggr_method: 'LVA' n_layers=2 vs_weights=[4,1] dis_hidden_size_list=[16,8,4] filter_hidden_size_list=[128,64] activation=leakyrelu fair_weight=0.1 gcn_n_layers=2 hidden_channels=32 gcn_dropout=0.2 gcn_act: relu train_epoch_interval=5 weight_decay=0.0001 threshold={rating: 3.0} |
| FairGo_GCN(LBA) | pretrain_model_file_path= Your GCN Pretrain File Path aggr_method: 'LBA' n_layers=2 dis_hidden_size_list=[16,8,4] filter_hidden_size_list=[128,64] activation=leakyrelu fair_weight=0.1 gcn_n_layers=2 hidden_channels=32 gcn_dropout=0.2 gcn_act: relu train_epoch_interval=5 weight_decay=0.0001 threshold={rating: 3.0} |
| FairGo_PMF(WAP) | pretrain_model_file_path= Your GCN Pretrain File Path aggr_method='WAP' dis_hidden_size_list=[16,8,4] filter_hidden_size_list=[128,64] activation=leakyrelu fair_weight=0.1 train_epoch_interval=5 weight_decay=0.0001 threshold={rating: 3.0} |
| FairGo_PMF(LVA) | pretrain_model_file_path= Your GCN Pretrain File Path aggr_method='WAP' vs_weights=[4,1] dis_hidden_size_list=[16,8,4] filter_hidden_size_list=[128,64] activation=leakyrelu fair_weight=0.1 train_epoch_interval=5 weight_decay=0.0001 threshold={rating: 3.0} |
| FairGo_PMF(LBA) | pretrain_model_file_path= Your GCN Pretrain File Path aggr_method='WAP' dis_hidden_size_list=[16,8,4] filter_hidden_size_list=[128,64] activation=leakyrelu fair_weight=0.1 train_epoch_interval=5 weight_decay=0.0001 threshold={rating: 3.0} |
| NFCF | load_pretrain_path= Your Pretrain Model Path dropout=0.2 fair_weight=0.1 mlp_hidden_size=[128,64] weight_decay=0.000001 threshold={rating: 3.0} |